Key Takeaways

  • Customer-service AI investment is rising faster than demonstrated financial returns, making operational discipline as important as chatbot deployment.
  • Typewise has launched Nova, an AI Operator designed to monitor, test and improve customer-service AI systems continuously.
  • The company's proposition addresses a practical barrier for smaller teams: the ongoing work of maintaining prompts, knowledge, workflows, evaluations and guardrails.
  • Early customer results cited by Typewise show high autonomous-resolution rates, but these remain vendor-reported outcomes rather than independently validated benchmarks.

Customer-service leaders are no longer deciding simply whether to deploy AI. The harder question is how to operate it after deployment. Virtual agents, agent-assist tools, knowledge systems and automated workflows can each produce useful results, but maintaining their accuracy as policies, products and customer needs change creates a continuing operations burden.

That challenge arrives as spending accelerates. Gartner reports that service leaders are investing a median 12% of their 2025 budget in AI (the highest share among business functions) yet only 24% report positive financial returns across their AI use cases, according to AI Reshapes Customer Service Strategy as Leaders Rethink Staffing, Skills, and Service Models: Gartner January 12, 2025. The gap suggests that buying AI capabilities and running them reliably are different undertakings.

From AI Tools to AI Operations

In its September 10 press release, Typewise introduced Nova, an AI Operator intended to build, operate and continuously improve an organization's AI customer-experience team. Rather than serving as another customer-facing bot, Nova is positioned as the layer behind AI agents: monitoring performance, testing changes, updating knowledge, tuning instructions and managing workflows and guardrails.

That distinction matters in a market where disconnected deployments are common. The company's 2026 Agentic AI Index, based on a March 2026 survey of 207 customer service agents across the US, UK and Germany, found that 81% of customer service teams still operate AI as a collection of disconnected tools rather than as a working system.

The operational task is not trivial. A team needs to determine whether responses remain correct, whether a changed policy has reached every customer channel, when an automation should hand off to a person, and whether a new workflow performs better than the prior one. Those are functions that large enterprises can assign to specialized teams; smaller organizations may find that a pilot becomes difficult to expand once the maintenance work becomes clear.

Why Resolution, Not Deflection, Is the Test

The customer-service AI market is growing around the promise of faster, lower-cost support, but the relevant measure is not merely whether a bot responds. Gartner forecasts that conversational AI will reduce global contact-center labor costs by about $80 billion in 2026, while automated interactions are expected to reach roughly 10% of total contact-center volume by 2026. Automation is therefore consequential, but it still represents only part of the service workload.

According to Typewise, Nova's agents resolve requests across email, chat and WhatsApp, rather than steering customers only to help-center articles. The platform can also handle pre-sales questions and product guidance, according to the company's press release, and integrates with shared inboxes, Intercom, Zendesk and Shopify. The company says the system transfers a conversation to a human when needed.

HealGreen, a Hamburg-based telemedicine platform connecting more than 120,000 patients with doctors and licensed partner pharmacies, is among the early customers cited by Typewise. According to the company, HealGreen moved from contract to live in days; its AI agents now handle 70% of incoming inquiries across email, chat and WhatsApp, and fully resolve 75 to 85% of those interactions, depending on channel and intent.

"Once you offer WhatsApp as a channel, clients naturally expect more than just inbound messaging, they want a seamless experience. We needed a platform that could orchestrate all our channels intelligently. The ability to connect our existing systems and databases directly with Typewise was a game-changer. Nova, the AI Operator, made the onboarding incredibly fast, allowing us to go live in days." Hendrik Knopp, CEO, HealGreen

The Case for Continuous Controls

The promise of an AI operator is not that customer service can run without human accountability. It is that AI can assume more of the maintenance workload while teams retain authority over what systems are permitted to do. That approach is particularly relevant as AI systems take action across channels and enterprise data sources.

The company says Nova allows teams to begin with AI drafting replies alongside employees and then assign trusted request types to automation. It says Nova evaluates its own output, tests changes and adjusts over time. That design raises an important buyer question: how are the system's evaluation criteria set, audited and overridden? Organizations should establish their own acceptance thresholds, escalation rules and governance processes rather than treat autonomous optimization as self-validating.

Customer sentiment reinforces the need for controls and human escalation. A 2026 Gartner survey found that 87% of customers want companies using generative AI for customer service to offer a human-agent option, as reported by Gartner Survey Finds 87% of Customers Say Companies .... For service teams, this makes the quality of handoffs as important as self-service containment.

"Every company that deploys AI agents creates a new operations job: someone must monitor them, update their knowledge, test changes and decide when they are ready for more responsibility. Nova is that 24/7 operations team. Companies delegate the operation to AI itself while keeping control over what it is allowed to do." David Eberle, CEO and co-founder, Typewise

Customer Results Need Operational Context

Typewise also points to its work with international consumer-health company Beurer. According to the company, 65% of incoming inquiries are resolved fully autonomously, with a resolution rate above 90% on those cases. For interactions still handled by people, average handling time fell from 12 to 6 minutes.

Those figures illustrate why buyers should separate autonomous-resolution rates, resolution quality and human-agent efficiency when evaluating an AI platform. A high automation rate can be unhelpful if customers are forced into repeated contacts; conversely, a lower automation rate may be valuable if it routes complex cases quickly to skilled staff. First-contact resolution, repeat-contact rate, escalation quality and customer effort should be evaluated alongside handling time.

The company says customer data can be hosted in Europe or the United States, with European customer data remaining in Europe. It also says it uses zero data retention and does not use customer data to train any model. At HealGreen, Typewise says Nova proactively flags and improves GDPR-critical settings. Those claims should be assessed against an organization's own data-processing requirements, contractual terms and security controls.

Common Questions

Does Nova replace customer-service agents?

Typewise positions Nova as operating the AI system behind customer-facing agents, not simply replacing employees. Its stated model includes handing conversations to people when automation is not appropriate, while also reducing the maintenance work associated with AI operations.

How should a buyer evaluate autonomous-resolution claims?

Buyers should ask how "fully resolved" is defined, whether the interaction avoided repeat contact, and how quality is measured across channels and intents. The HealGreen and Beurer metrics cited by Typewise are company-reported results and should be validated against the buyer's own customer mix, policies and escalation requirements.

What systems can Typewise connect to?

According to the company's press release, Nova works with shared inboxes, Intercom, Zendesk and Shopify, and can support email, chat and WhatsApp interactions. Prospective customers should confirm the specific data, workflow and identity integrations required for their environment.

What Comes Next

The next phase of customer-service AI will be defined less by who can deploy a conversational interface and more by who can sustain safe, accurate automation across changing business conditions. Nova enters that operational layer with an outcome-based model in which a full resolution counts as one credit, a partial resolution as a half credit, and an unresolved request costs nothing. Whether that model delivers durable value will depend on measurable resolution quality, disciplined governance and customers' willingness to keep humans accessible when the automated path falls short.